Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 10 Jun 2002 00:57:44 -0400 (EDT)
From:      Robert Watson <rwatson@freebsd.org>
To:        Luigi Rizzo <rizzo@icir.org>
Cc:        ipfw@freebsd.org
Subject:   Re: New ipfw code available
Message-ID:  <Pine.NEB.3.96L.1020610005651.89066B-100000@fledge.watson.org>
In-Reply-To: <20020608201909.A41807@iguana.icir.org>

Next in thread | Previous in thread | Raw E-Mail | Index | Archive | Help
Sounds very cool indeed.  However, the usual question when hard-coded-ness
is traded for flexibility is: what's the performance like?  Do you have
any performance measurements you could tell us about in the before/after
scenarios?  You mention 'faster' as well as 'flexible', which bodes well
:-).

Robert N M Watson             FreeBSD Core Team, TrustedBSD Projects
robert@fledge.watson.org      Network Associates Laboratories

On Sat, 8 Jun 2002, Luigi Rizzo wrote:

> [Bcc to -current because it is relevant there as well -- sorry for the
> crosspost]
> 
> 
> Hi,
> over the past 2-3 weeks I have done an extensive rewrite of the
> ipfw code (userland + kernel) in an attempt to make it faster and
> more flexible.
> 
> The idea (which I discussed a few times on the mailing lists) was
> to replace the current ipfw rules (macroinstructions) with a set
> of microinstructions, each of them performing a single operation
> such as matching an address, or a port range, or a protocol flag,
> etc.  -- much in the spirit of BPF and derivatives -- and to let
> the userland front-end compile ipfw(8) commands into an appropriate
> set of microinstructions.
> 
> There are several advantages in using this technique: first of all,
> instructions are typically shorter and faster, because the former
> code had to check for the presence of all the possible options in
> a rule, whereas the new one can simply do just the things that are
> required --  e.g. an instruction like
> 
> 	allow ip from 1.2.3.0/24 to any
> 
> translates to a couple of microinstructions (whose complete
> implementation is below the instructions themselves):
> 
> 	O_IP_DST 
> 	    if (((ipfw_insn_ip *)cmd)->addr.s_addr ==
> 		(dst_ip.s_addr & ((ipfw_insn_ip *)cmd)->mask.s_addr))
> 		    goto cmd_match;
> 	    goto cmd_fail;
> 
> 	O_ACCEPT:
> 	    retval = 0;     /* accept */
> 	    goto accept;
> 
> 
> But there is a lot more -- the instruction set is easily extensible,
> and without backward compatibility problems.  Furthermore, you can
> build (and I have already implemented them) more complex rules by
> assembling microinstructions with OR and NOT operands. I.e. you can write
> something like:
> 
> 	pipe 10 tcp from 1.2.3.4 or 1.2.3.7 or not 1.2.3.0/28 21-25,1024-4095 \
> 		to any in recv ed0 or recv fxp1 or recv dc0 uid 35 or uid 50
> 
> You get the idea... 
> 
> I have a fairly complete version of the above  code at the moment,
> which is only missing a small set of functionalities
> (ip/tcp flags matching, "log" and fixing hooks to the stateful
> code). However the glue to implement all the missing pieces is
> already there, it is just a matter of adding a few lines of code
> and testing things.
> Other than that, the code is meant to be fully compatible with the
> old syntax so you will not have to rewrite your existing rulesets.
> 
> I have put a preliminary snapshot of this code (for CURRENT) at
> 
> 	http://info.iet.unipi.it/~luigi/ipfw5.20020609.tgz
> 
> It replaces the following files from a recent (2002/05/14) version of -current.
> 
> 	sys/netinet/ip_dummynet.c
> 	sys/netinet/ip_fw.c
> 	sys/netinet/ip_fw.h
> 	sbin/ipfw/ipfw.c
> 
> I would be very grateful if someone could have a look at the
> code, maybe give it a try, and see e.g. how it compiles your
> typical ruleset and whether the new extensions can make your
> ipfw rulesets simpler.
> 
> Feedback welcome, both on the architecture and on the implementation.
> 
> NOTE: if people wonder why I did not use BPF and reinvented the wheel:
> the keyword is "backward compatiblity" -- i thought it was a bit too
> complex to compile the existent ipfw syntax into BPF, especially because
> BPF at least as far as i know does not handle UIDs, and GIDs and
> interface matches and different "actions" than match or not match,
> so i would have had to extend the code anyways, at which point i
> thought I could as well write my own microinstruction set...
> 
> 	cheers
> 	luigi
> -----------------------------------+-------------------------------------
>   Luigi RIZZO, luigi@iet.unipi.it  . Dip. di Ing. dell'Informazione
>   http://www.iet.unipi.it/~luigi/  . Universita` di Pisa
>   TEL/FAX: +39-050-568.533/522     . via Diotisalvi 2, 56126 PISA (Italy)
>   Mobile   +39-347-0373137
> -----------------------------------+-------------------------------------
> to 
> 
> 	thanks
> 	luigi
> 
> To Unsubscribe: send mail to majordomo@FreeBSD.org
> with "unsubscribe freebsd-ipfw" in the body of the message
> 


To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-ipfw" in the body of the message




Want to link to this message? Use this URL: <http://docs.FreeBSD.org/cgi/mid.cgi?Pine.NEB.3.96L.1020610005651.89066B-100000>